S2. Ep37 - Can You Really Build an App in a Day? We're Calling BS

Katie (00:00)

Hello, welcome to another episode. Hi, hello, I'm Katie. And as always, I have Noel with me today. Hi Noel, how are you doing?

Noel (00:09)

I'm doing absolutely fabulous again this week. How are you doing?

Katie (00:13)

Yeah, all good, thank you. Really looking forward to this week's episode.

But before we even think about jumping in and discussing this week's topic, we always start the podcast episode with any news, any updates from the world of AI and automations. So Noel, tell us, what updates do we have this week? Do we have any updates?

Noel (00:47)

We do, we have lots of updates. And I also feel smug, because we're recording this on a Wednesday and one happened yesterday on a Tuesday evening, like they usually do. So brilliant, I can capture it this week.

Noel (01:01)

So last week we talked a lot about the AI slowdown, and how models are potentially going to come out slower in the future.

Katie (01:06)

Yes.

Noel (01:12)

Well, that didn't last long, because we've had four big model releases this week since recording the episode. I don't know, their definition of slowdown is probably different to mine, but there we go.

Noel (01:28)

So just to go over them very quickly. GPT-6, that was Astra, which is OpenAI's biggest, best model. This week they released GPT-6 Sol and Luna. Those are pretty good. They're currently working out about 50% cheaper versus the 5.6 variants. So if you're running those in your automations already, then I think it'd be well worth switching those out and getting that extra saving. Other than that, it's very much the same use case. They've got smarter, they're getting cheaper, that's the main upshot.

Noel (02:10)

We also had Grok 4.7. Although this one has actually stayed at the same price, and there's only a small uplift. I don't really get that excited about Grok models, if I'm perfectly honest.

Katie (02:24)

No. Okay, personal preference, that's fine.

Noel (02:30)

Yeah. Someone listening might use Grok, so that could be helpful. So that's what we'll go for.

And then last night, Anthropic released Opus 5.5. This one is actually more capable than Fable 5.1, which was their top model.

Katie (02:49)

Mm-hmm.

Noel (02:49)

And this one is 20% cheaper in terms of tokens in, tokens out, but it's actually more efficient, so your savings could go up to about 40% cheaper. It's a big jump from Opus 5 in terms of knowledge, capabilities and intelligence. So quite nice to get a bit of a saving and a bit of a boost at the same time.

Katie (03:17)

Yeah, that sounds good.

Noel (03:19)

So those are the main model updates. But there's actually another model that came out last week which has got me very excited. And it isn't a chat model, so we can't just have a conversation with it like we can with these other models. This one is there just to do certain specific tasks.

Noel (03:50)

So you could say to this AI, which is called Jev, look, here is an email. What is it about? Does this need to be routed to sales, does it need to be routed somewhere else? And it can just look at that particular email and then make a snap decision. It goes, right, that needs to go to the sales team. And then your automation can route it in that direction.

Noel (04:20)

Now the thing that's got me really excited about it is how fast it can do this. At the minute we're looking at about half a second per thing that goes through it to get a decision, and it costs pretty much nothing to run. So if any businesses are using any sort of AI agent or large language model to do some sort of categorisation or classification of text, then I would definitely go off and check out Jev, because it's got no hallucinations. It will just do the job in a super fast time and be super cheap.

Noel (04:50)

With the traditional models, you've got to wait for them to do the reasoning, you've got to wait for them to think about it, and then they create a really long response about it, and then it goes, right, yeah, it should go over there to the sales team. So it's going to be a lot faster and cheaper.

Katie (05:05)

Okay.

Noel (05:06)

As a real use case as well, I've built a couple of things with Jev already this week. With my agent builder Clyde, he has access to lots of different tools which he can go off and use. But he doesn't need to know about all of them in order to do a simple job. So I have Jev in the middle who's filtering out all the tools and then goes, right, well, based on what Noel's asked for, Clyde only needs access to these eight tools. So it's reduced my token usage on those tools by over 90%.

Noel (05:45)

So that's what I mean, it's super clever, super smart, super fast, and it can make those decisions for you. But I also built a LinkedIn filter as well, in a Chrome extension. It works that fast, it can hide posts and highlight posts in real time as you're scrolling through. So I built a little extension for myself to bring the things that I want to read to me first, and then stuff that I'm not interested in, just hide it. It can do all that in real time. So that's why I'm really excited about it, if you can't really tell.

Katie (06:20)

Yep, we can tell.

Noel (06:26)

Good. Yeah, it's well worth checking out.

Katie (06:31)

Okay, any further updates?

Noel (06:36)

No, that was it for this week. I'm sure there'll be even more next week. Who knows, we'll find out.

Katie (06:44)

Okay, so going into this week's topic: can you really build an app in a day?

So I said to Noel last week, I am being hit by ads all over social media from different people. And I'm not signed up to many business newsletters anymore, I did a big unsubscribing cull because my inbox was getting to the point where it was unmanageable, but they've even entered my inbox, my sacred inbox. And they all claim and talk about very similar things: how you as a business owner can build an app in one day.

Katie (07:50)

So these aren't aimed at technical business owners. These are mostly service-based business owners who don't have a team, solopreneurs. And I want to discuss it, because me personally, I think it's BS that you can do that. But Noel, you're the expert, and I'd love to hear your take on it.

Noel (08:19)

Yeah. So for me, I've been building apps and all sorts of stuff for businesses that I run and for other people for a number of years now. So I've been at it a while when it comes to AI and coding agents. So could you build it in twenty four hours? I think the answer's yes and no.

Katie (08:43)

Okay, and I feel like I should add a bit more clarity. A lot of these ads and emails that I'm getting are saying, you don't even know what type of app you might need for your business. So you don't even know what you're going to build for your business, and then you're going to come up with that idea, and then you're going to build it, and then you're going to launch it, all in a day.

Katie (09:15)

So I just wanted to clarify that that is what I'm seeing a lot of, and that's why I'm calling BS, and that's why I really wanted to discuss it on the podcast.

Noel (09:36)

I guess, are these pitched as internal or customer facing tools, or a bit of both?

Katie (09:42)

Customer.

Noel (09:43)

Ooh. Then I'm going to change my yes or no to no.

Katie (09:48)

Okay, that's it, it's just a hard no. Okay, end of podcast, everyone.

Noel (09:53)

Yeah. But obviously I'll give all of my reasoning as to why. It doesn't mean you shouldn't, or don't take this as don't try it. By all means, these days it's quite easy to get started, and you can start building stuff. But should you build and deliver something customer facing that's quite technical in a day? I think for any business owner, if we read that, we're a bit like, that seems risky.

Katie (10:27)

Yeah. So do you think it's a bit of a marketing gimmick on behalf of the businesses that are saying, come and build an app with me in a day?

Noel (10:43)

100%, yeah. Because there's quite a lot to it. It's not just making something functional and making it look nice, that's just one part of the problem when it comes to building an app. But it's the delivery of it. It's all the security stuff, like how are people going to access it. There are lots of things.

Katie (11:05)

Okay, so let's go through this then in a bit more detail. Let's do a step by step. So what, first of all, is actually involved in building your own app that is for your customers?

Noel (11:28)

So the first thing you need to do is sit down and start writing down some requirements. What's the aim of what you're trying to build? Is it something that's going to be useful, for a start? If it doesn't make it through the useful gate, then stop and think about something else.

Noel (11:46)

If no one's ever going to use it, then there's no point going any further. But what I always do is I sit down and work out, well, what's the main aim of this particular app that I'm going to build? How am I going to go about that?

Noel (12:10)

There are things like, how are users going to log in and sign in, for a start? If it's an online tool, first of all you need to figure that bit out, because you can't just have everybody having access to loads of data. It just becomes a mess, and then people can access someone else's data, and then you get into all kinds of trouble with GDPR and HIPAA, and we don't want to get into that realm.

Katie (12:35)

No, thank you.

Noel (12:37)

No. So once we've worked out what it's going to do and the rough flow, from someone logging in to how they then interact with the app, what data are you going to show them within the main screen? What's it going to say? How is it going to work? I would sketch all of that out. It doesn't need to be super detailed, but just a very rough, this happens, then they go here, then that happens, and then they get the end result at the end. What is that end result?

Noel (13:12)

But once you have that flow, for me the next thing is data. Everything's going to need something stored somewhere in order to read it and send it back, so you can see it on the screen. So my go-to is always Supabase, because it's really cheap to get started. You can have a free account, you can set up a database in there for free.

Noel (13:55)

Which for some people, I'm sure yourself included, Katie, would sound absolutely horrifying, that you would have access to a database. But the coding agents that we use, they can create all of the code and create all of those tables for the data and the relationships for you. So you don't really have to worry about that.

Noel (14:10)

And also within Supabase is the user authentication. They cover the entire sign in, sign up process as well, and it can link into your app. So that kind of puts the security thing half to bed. There's more that needs to be worried about, but that's the main bit.

Noel (14:24)

So those are the big things that I worry about when I first start building. The flow, then the data, and the user accessibility, how they get in and out.

Katie (14:36)

Yeah, okay, so there's quite a lot to think about.

Noel (14:42)

Yeah. And I guess, to take it back another level, it's also where is this app going to live? Is it going to be a browser based app? Is it going to be an iPhone or Android app? So you need to figure that bit out as well. Where are your users going to access it? Is it going to be a little link on your website that then goes off onto another page which hosts the particular application? So lots of things like that to consider.

Katie (15:09)

I guess as well, with an app, you actually have to keep it updated and make sure any of the components that you've used to create the app are updated as well on a regular basis.

Noel (15:30)

Yes. So I think, for the most part, most of the code which AI would produce would use libraries, in the coding sense. So it would grab that code from a library and then use that to build the app. So I would say you would eventually need to go through and do it, but I wouldn't expect to be doing it every month and thinking, I'm going to have to update the code every month. It could be every couple of years that you would need to go back and revisit it, potentially.

Noel (16:06)

But if you've got AI in there as part of your offering and your tool, then yes, you would probably need a method to update models, so your customers' experience is moving as AI develops. So when I have AI in my apps, I always have the option of going, right, well, which model does everyone have access to today?

Noel (16:30)

We're building one at the moment for a business, and I went in there straight away today and changed Opus 5 to 5.5. So now every user goes through and they have that updated model, which is good for us because it's cheaper, but good for them because they get the better outputs as well.

Katie (16:49)

Yeah, okay. So do you think that every business actually needs to have an app?

Noel (17:02)

No, I don't think so. I think if you look at it long and hard, you don't really need to. Just because you can doesn't mean you should.

Katie (17:14)

Which you often say on this podcast as well. Just because you can doesn't mean you should.

Noel (17:20)

Yeah.

Katie (17:20)

So do you think it's almost like people are trying to hype people up and saying, you could have an app for your business, and it's going to be a better experience for your customers, but actually it could just be that shiny object? It might not really be necessary, and it might mean a lot of hard work for you and actually not really have that much of an impact.

Noel (17:58)

Yeah, it is tricky when you start to look at it that way, I guess.

I guess the biggest moat that we used to have was cost when it comes to building apps. You would have to go off and hire a developer at an insane price if you wanted something complicated with lots of data and security. As soon as you start mentioning security, the price always seems to shoot up.

Noel (18:40)

So now, I'm not saying you should never go off and hire a developer to do a job, but it gets to a point where you can start creating a minimum viable product within a week or so, on your own with AI models. The cost reduction in that alone is massive. You're probably saving yourself a hundred thousand pounds or dollars, just by doing that.

Noel (19:10)

It's enticing. But it depends on where your business is. If you're heavily regulated and things go wrong, then you could get fined more than what it would have cost to get somebody in to do it properly. So I would weigh those risks as well, if that applies in your use case.

Katie (19:26)

Okay. So thinking about all of that, what would be the cost involved in building your own app and then running it?

Noel (19:39)

So it's quite cheap when you're using AI. We can create stuff using either our ChatGPT or Claude or Gemini subscription. So we're already paying for that AI usage, so you can class that as almost technically free, I guess, in some senses. You were going to use it anyway.

Noel (20:10)

But there are some subscriptions. I've mentioned Supabase, which is free, but if you grew to a certain size, then it's twenty five dollars a month. So not a huge amount.

There are also platforms where you then need to host the app that you've built. So if it's going to be a web app, for me I use Vercel, and that's free to start off with. You don't even have to buy a domain, they'll just give you a little cheap domain which you can use so you can view it online.

Noel (20:50)

I guess it all depends what you're going to connect into your app. Is there a third party where you're going to get data from, or use their data to do stuff? Is that going to cost you? But for the large part of it, you could get away with mostly free to get you to that minimum viable product.

Noel (21:05)

As and when you move up and get users in, that's when your costs will start to go up. You'll start using more of your usage, you'll have a bigger database, you'll have lots of people going on your website which increases the costs on Vercel. So it does scale up over time, but it's not massive, I would say.

Katie (21:22)

So, just with my thinking as you were talking, what happens when, say, 200 people are using the app all at once?

Noel (21:36)

That is also another great consideration. Once you've got that minimum viable product running and it works for you, which is fantastic, that's problem number one. Problem two, like you say, is scale after that.

Noel (22:00)

So what I would do is I always go back to the AI models and say, well, where am I going to start hitting bottlenecks? How many users could I have with this particular setup before I start hitting problems? And it can quite easily go off and figure that out for you, and say, well, you can easily get to a thousand before you need to start worrying.

Noel (22:20)

Well, a thousand people paying maybe twenty to fifty dollars a month, that's a great problem to have. You would then have the capital to fix that problem, and increase your usage on any particular application which helps you run it. So that's how I would look at it.

Katie (22:34)

Okay, yeah, that makes sense. And I guess we haven't spoken about how to actually build these apps, like what tools you would be using. And the cost of those as well would be really interesting to know.

Noel (22:56)

So to get started with the tools, like I've mentioned, you would need ChatGPT, Claude, Gemini, something like that. They have their own versions of coding apps, so you've got Claude Code, Codex, Gemini have got their own, I don't think they give it a name. But that's the thing which will create that code for you. Most of that's within a subscription.

Noel (23:25)

You can get through your usage credits pretty quickly. So if you're on one of the middle plans on their platforms, then you are going to run out of usage very, very quickly. So getting this done in 24 hours is going to be incredibly constrained by what account you have access to. If you're only on a ChatGPT Pro, that's not going to happen in a day. Forget that. You're going to hit your usage limit.

Noel (24:00)

And then you're going to be in AI jail for a couple of hours or a day, and then the next day you can restart. So that's the cost side of things sorted out. But like I said, it does all depend on what it's going to be, the databases and the data it has access to. So I would budget at least a hundred dollars a month on subscriptions to get you to a point of having something that does work.

Noel (24:40)

For me, I always use the same tool set. I always use Claude Code, that's my favourite. I will always use Supabase as the database, because I can get stuff for free and it's really simple to use. And Vercel for the front end, for having it as a live website.

Noel (25:00)

When it comes to scaling and usability, you should also probably look at having what we call a backend server, which does all of the heavy work. So if there's any data processing that needs to sit somewhere else. For me, I use an app called Railway, which is like five dollars a month. And you can have multiple different servers running on that five dollars a month, and we haven't hit the cap on that.

Noel (25:20)

I've got all sorts on there. I've got Clyde on there, I've got some other servers running on there all at the same time, and I still haven't hit my five dollar limit. So there's lots we can do with that. But yeah, at least a hundred, hundred and twenty five dollars a month, I would put it down as.

Katie (25:35)

Okay. And what would you say is... because Noel, you've built so many apps, not just for yourself but also for clients. What's one thing that people forget or don't think about when they are building an app?

Noel (25:56)

I think the big one that always seems to slip through the gaps is security. I won't say it fully slips through the gaps, but it's there as a thought, it's never designed first. It's always kind of an afterthought.

Noel (26:20)

Especially with people that I've worked with. They're kind of like, yeah, make this, and it looks great and fancy, and then they're like, hang on, hold on a minute, how secure is this? Who has access to the data? And it's like, right, okay, well, I've already been working on that in the background, that's fine. But to them, it's a bit of an afterthought, which is a huge risk in itself.

Noel (26:50)

I also think most people would forget about the user experience side of things. So I always try and segregate things. The way my brain works, I have to write that flow down, or have it written down somewhere, so it makes sense to me and I can see that customer journey.

Noel (27:15)

Most people would just go off and go, right, Opus, go make me this thing, and then Opus will figure out that user flow for you. It can definitely do it. But is that actually right for you and your customers? Is it going to make sense?

Noel (27:30)

Because sometimes we can build stuff with it and go, hang about, that's not what I was expecting. I want them to go left instead of right in this routing, and then it can figure it out. But I'd always keep an eye on user experience, because if it's unusable, no one's ever going to use it.

Noel (27:46)

They'll come in, they'll pay for it and then go, I've no idea what's going on, there's stuff everywhere, and then cancel and move on. So I'd be very wary of that.

Katie (27:57)

Okay.

Noel (27:58)

It's all just got to make sense. Some people will come to me with an idea and they haven't quite fully thought about it to the end degree. It's kind of like a spark, and then they've gone, right, we need to build this. But they haven't sat down, gone through those requirements and done that groundwork beforehand, which is really, really important, because we can ground the AI coding agent with those requirements and that sort of thinking right from the get-go. And that's a big, big thing.

Noel (28:40)

And with that, obviously we can sit down and write all that out in a document ourselves, but you can also get AI to help you do that.

Noel (28:55)

The way I usually start is I'll start a fresh chat session or a fresh project in Claude, and then I would say, look, this is the kind of thing I'm looking at doing. It'll go off, it'll do research, it will look up other apps. What are they doing? How are we going to differentiate between them? What's our unique selling point? And from there it can then build you a product design document.

Noel (29:20)

So we can say, go off, create that requirement spec for me, and it'll create it for you. And then as soon as you've done that, you can hand that off to Claude Code and go, look, this is what I want building, this is the document to build against. And then that gives it a really good foundation to get started.

Katie (29:40)

Okay, so I guess then, where do we land? Can you build an app in a day?

Noel (29:54)

No. A customer facing one, no, definitely not. That would be a lot of work to get up and running.

Katie (30:00)

Okay.

Noel (30:01)

I guess it also depends on how complicated your app is going to be. It could be quite simple, I don't know. But most people and most businesses will want lots of different features and things that the user can do. So I would definitely put that in the no camp.

Noel (30:24)

But if you're building something that's going to be for you and only you ever use it, it could be a particular tool or calculator or something quite basic, then yeah, you could do that in an hour. No one's going to see it, it's only for you, so that won't take you long at all. But customer facing, no.

Katie (30:44)

Okay. So if someone thinks, well, I would like an app for my business, and they've seen all of these "you can build an app in a day" ads, and then they go to an agency or someone like you, Noel, and you quote them, I don't know, a couple of weeks. How would you manage their expectations for timelines and of course for pricing?

Katie (31:20)

Because it is really time consuming. This is why I wanted to have this conversation, because obviously I've been around when you've been building apps and working on apps, and I've said to you, Noel, what are you working on today? And you're like, this app. And it's been seven or eight days that you've been working on the same app.

Noel (31:43)

Yeah. So like I said, there are lots of considerations when it comes to the length that I would normally quote for a project. What I always find is that the scope always creeps, for a start. So there's always going to be that buffer where I'm like, right, yes, I could build it in a couple of days, but they're going to see it and then come back with, seeing as it does this, can it do this instead, or do that?

Noel (32:15)

It's one thing thinking about it, it's another thing seeing and using it, and then going, right, well, we need to change tack on this, we need to go this way instead of what we were doing. So that's always a consideration and things like that move the timeline and extend it.

Noel (32:40)

So whilst people say that it can be done in a day, if you want to do it properly, don't rush it. That's the big point. There's no point rushing something out just for somebody to look at it and then it breaks within five minutes. Because then you've got the support element that comes behind it, people asking questions, going, well, I've got this bug, it didn't work, how do I fix that? You've got that cycle.

Noel (33:10)

So what I also build in is a testing phase. So you'd be like, right, we've got to this point. Does it work? Does it give us what we need? Do we need to add anything else in? Does it functionally work? We can try and stress test the app. So there are lots of considerations for that sort of customer facing app.

Katie (33:33)

Yeah. And I guess as well, business owners do want the app to look nice. They want it to look branded. And I feel like even just getting it branded and looking nice is definitely going to take more than a day.

Noel (33:51)

Yes, definitely. Everyone has their own style and how they would like it to look. But that always comes from viewing it. You could give the brand guidelines to the AI and say, these are my fonts, these are my colours. But it's not until you see it that you go, no, no, this looks terrible, we need to change this. And that can take quite a bit of time to get right.

Noel (34:15)

It's really important to get it right, because it's got to look consistent with everything else that you do.

Katie (34:22)

Yeah, of course. People have got to think, well, I'm logging into this app, I might be even simply giving an email address. They want to make sure that it is actually trustworthy. And in order to do that, it has to have that recognition of your brand, because they already trust your brand.

Katie (34:50)

So then when they download the app, if it looks completely different, or if it looks a bit dodgy or a bit sketchy, they're going to be like, is this actually the right one? Because this looks nothing like the business that I know.

Noel (35:05)

Yeah, you don't want something that looks like Windows 95. Or with AI, it will always create stuff in dark mode normally, first. So if your website's bright and colourful and then your app is dark mode, purples and blues, you'd be like, hmm. It's questionable.

Katie (35:25)

Yeah. Okay, that's really interesting. So yes, you can build an app in a day, but also no, you can't build an app in a day.

Noel (35:35)

Yeah. It's the old saying, just because you can doesn't mean you should.

Katie (35:41)

Yeah.

Noel (35:42)

But I would always say to people, just take your time with it. We don't need to rush it out. Just because someone else has rushed something out that's there in the wild today doesn't mean you have to. Theirs is probably going to be full of issues, and they're going to be working twice as hard to fix issues while they've got paying customers. So just get it right first time if you can.

Katie (36:11)

Okay. Well, we would love to hear from you if you think, should I build an app for this? Email us, hello@makeautomations.ai, and we'll answer, should you build an app for this?

Katie (36:30)

But we always love hearing from you. If you've got any thoughts, feedback or questions from this week's podcast episode, please do email us. It's hello@makeautomations.ai and we will get back to you. We might even give you a shout out on the podcast as well.

But thank you so, so much for listening to this week's podcast episode. We hope you have enjoyed it, and we will catch you next time for another episode very soon.